Testing performance of a 1D directional Gaussian blur: single pass vs. multiple passes.
This version uses a true Gaussian kernel. Applying the Gaussian blur `P` times with a standard deviation (`σ`) per pass adjusted as `σ_pass = σ_target / sqrt(P)` yields a result mathematically equivalent to a single pass with `σ_target`.
Note: The visual output should be identical (within precision limits). The total number of texture fetches per pixel will differ between modes (Single: `2*ceil(3*σ_target)+1`, Multi: `P * (2*ceil(3*σ_pass)+1)`).